Introducing

Luxonis
OAK 4

The world’s first fully self-contained AI vision platform. Compute, perception, and cloud management unified into a single system-bringing physical-world AI fully to the edge.
Scroll to explore
Scroll to explore

Industry leading AI Hardware

Each variant engineered for specific use cases, all powered by the same cutting-edge platform.

OAK 4 S

Anything AI
US $749

Ships before Christmas

OAK 4 S
TOPs
52
CPU
6-core CPU + GPU
Connectors
PoE and USB
IMU
Yes
Dot Projector
No

OAK 4 D PRO

Active Stereo
US $949

Ships before Christmas

OAK 4 D PRO
TOPs
52
CPU
6-core CPU + GPU
Connectors
PoE and USB
IMU
Yes
Dot Projector
Yes

OAK 4 CS

Swappable Lens
US $899

Ships January 2026

OAK 4 CS
TOPs
52
CPU
6-core CPU + GPU
Connectors
PoE and USB
IMU
Yes
Dot Projector
No

Built to Last

OAK 4 is engineered for demanding environments. From construction sites to autonomous vehicles, these devices deliver reliable performance when and where it matters most.

IP67 Rating

Dust-tight and water-resistant for industrial and outdoor use-reliable in rain, dust, and wash-down environments without extra housing.

Extreme Shock & Vibration Resistance

Passes ISO19014 and IEC60068; engineered to withstand up to 2000 Hz vibration and 30G shocks for consistent performance on heavy machinery and mobile platforms.

Wide Temperature Range

Operational from –20°C to +50°C, enabling deployment in cold storage, hot factory floors, and harsh outdoor environments.

PoE & USB Connectivity

Flexible power and networking via Power-over-Ethernet or USB, simplifying installation across robotics, vehicles, and fixed infrastructure.

Management Platform

Luxonis Hub:
From Hardware to Platform

Luxonis Hub is the cloud device-management software that transforms OAK 4 into a complete vision system-centralizing deployment, updates, monitoring, and improvement from a single intuitive graphical interface.

Devices

Device management lies at the core of Hub Control, enabling you to efficiently manage all your connected devices from a single interface.

Snaps

Snaps are lightweight image captures taken from your OAK devices, primarily used for building machine learning datasets. They offer a simple way to capture and manage image data manually or semi-automatically.

Apps

Application Management enables users to deploy, manage, and monitor applications running on OAK devices connected to Luxonis Hub. Users can browse existing applications or upload new ones, restart or stop running apps, and install them on specific devices.

Model Zoo

Model Zoo

Luxonis Models is a curated collection of public AI models from the Hub. It offers a wide selection of models, making it easy to find the right fit for your project.

A New Age of
Stereo Depth Performance

Industry-leading depth-estimation models designed to deliver accurate 3D perception across a wide range of environments.

<1%

Depth error at close range (<2m)

20m

Maximum effective depth range

120FPS

Real-time depth processing speed

Neural Stereo Depth

Neural Depth combines traditional stereo matching with deep learning models trained with a diverse set of real-world scenes. This hybrid approach delivers exceptional depth accuracy even in challenging lighting conditions and with difficult surfaces.

Explore Real Point Cloud Scenes:
Point Cloud preview

To interact with the point cloud, click anywhere on the panel.

LENS

Luxonis Edge Neural Stereo

Proprietary best-in-weight class models supported in 4 different sizes.
Real-Time Performance

10 - 60 FPS

Hardware-accelerated inference for instant depth estimation.
Edge Computing

On-Device Inference

No cloud dependency for maximum privacy and speed.

AI Enhanced Perception

Explore our proprietary vision technologies that push machine understanding closer to how humans see the world.

Dynamic Calibration

OAK uses its patent-pending dynamic calibration system to heal its depth perceptions in case of drifts caused by extended use in extremely harsh environments.

Focused Vision

A prebuilt Luxonis application that uses model chaining on OAK 4 - detecting subjects, cropping intelligently, and running high-resolution inference only where it matters.

Multimodal Fusion

Seamless integration of multiple sensor modalities for enhanced perception and understanding.

  • IMU + RGB + Depth
  • Temporal alignment
  • Hardware synchronization
  • Calibrated outputs

Real-World AI Performance

Run state-of-the-art CV models directly on-device-delivering performance superior to standard Jetson Orin Nano systems.

Model

Input Size

Precision

FPS

Power

YOLOv6 - nano

512x288

INT8

830 FPS

2.7 W

YOLOEv8 - large

512x640

INT8

85 FPS

5.6 W

DeepLabV3+

512x288

INT8

340 FPS

4.6 W

YOLOv8-large Pose Estimation

640x352

INT8

170 FPS

4.9 W

Depth Anything V2

336x252

INT8

95 FPS

3.5 W

Paddle Text Recognition

320x48

INT8

625 FPS

2.4 W

DINOv3 Backbone

512x384

FP16

40 FPS

5.7 W

Luxonis supports TensorFlow, PyTorch, ONNX, and other model formats. For more models you can visit the Luxonis Model Zoo.

The OAK 4 Ecosystem

Powerful capabilities for advanced computer vision applications.

Powerful Libraries
Full ecosystem support for the most popular computer vision and AI libraries.
PyTorch
DepthAi
RTABMap
Opencv
OpenVINO
Open3D
YOLO-World
Real-time open-vocabulary object detection with zero-shot capabilities. Detect any object by simply describing it in natural language.
  • Detect custom objects without training
  • Natural language queries
  • Real-time performance at 60+ FPS
  • Works with any object category
Demo Applications
Explore real-world applications built with OAK 4. See the technology in action across various industries.
Open Vocabulary Object Detection
Open Vocabulary Object Detection
This example demonstrates an advanced use of a custom frontend. On the DepthAI backend, it runs either YOLOE (default) or YOLO-World on-device, with configurable class labels and confidence threshold - both controllable via the frontend.
Lossless Zooming
Lossless Zooming
This example demonstrates how to perform lossless zooming on the device. It will zoom into the first face it detects. It will crop 1080p frames, centered around the face. The example uses YuNet model to detect faces.
Age and Gender Detection
Age and Gender Detection
This example recognizes age and gender of all detected faces on the frame. The demo uses YuNet face detection model, crops the faces and then recognizes the age and gender of the person using Age-Gender model.
Segmentation Depth Crop
Segmentation Depth Crop
This example shows how to run the Deeplabv3+ model from our HubAI along with the StereoDepth node and crop the depth image based on the models output.
Hand Pose
Hand Pose
This example demonstrates how to build a 2-stage DepthAI pipeline for hand pose estimation. The pipeline consists of MediaPipe Palm detector and MediaPipe Hand landmarker.
Triangulation
Triangulation
This DepthAI example uses stereo cameras for face detection with YuNet, calculating 3D spatial coordinates via triangulation. It visualizes bounding boxes, keypoints, and disparity in real-time.

Why Choose OAK 4

Packed with features designed to accelerate your computer vision applications.

Compute + Sensors in One Device
A fully self-contained AI module combining RGB, stereo depth, IMU, audio, and edge compute in one synchronized unit.
Engineered for Extreme Durability
An IP67-rated, industrial-grade device built to operate reliably in harsh, mission-critical environments.
Cutting-Edge Depth Performance
Standard stereo depth plus on-device Neural Depth for accurate 3D perception even in complex real-world conditions.
Unified Into a Complete Vision Platform
Seamlessly integrated with Luxonis Hub for deployment, updates, and improvement—creating a single ecosystem from camera to cloud.

Need More Help?

Our dedicated team is available for technical support, business solutions, and more. Let us provide the help you need.